deepseek-v4.1POST
DeepSeek V4.1 Flash on two DGX Sparks: 2,135 tok/s on a 93K prompt
A user ran DeepSeek V4.1 Flash on two DGX Sparks, hitting 2,135 tok/s on a 93K-token prompt and loading 1M context in 27s, far above the old four-Spark vLLM setup's 48.5 tok/s.
Brad Mills
@BradMillsArchived 12 days agoOriginally posted 2026.10.19
Brad Mills @BradMillsOriginally posted 2026.10.19
deepseek V4.1-Flash on just two DGX Sparks. Two boxes reading a 93K token prompt at 2,135 tok/s, full 1M context loads in 27s. Beats the old four-Spark vLLM setup at 48.5 tok/s. Unified memory is a long-context cheat code.
x.com/bradmills/status/2112020312095084971Archived 2026.10.19 12:04